Author: Gabriel Horowitz Rossi
Graphics: Yarden Pri-Noy
Tech companies have made profound investments into AI Research. Does theoretical research reveal practical insights and solutions? What will be the ultimate impact of AI? Find out more about how this relates to our present and our possible future.
Investment, Investment and More Investment
In case you’ve been living under an incredibly heavy rock: artificial intelligence has arrived.
With years of investment and research behind them, companies like OpenAI, Google, and Apple have produced a wide variety of AI products, ranging from consumer-facing applications to powerful developer tools. These have become the transformative technologies that are positioned to reshape everything.
Alphabet, Google’s parent company, reportedly doubled their capital expenditures from 2025 to 2026, with somewhere between $175 billion to $185 billion being spent on AI infrastructure and development. Given these figures, it is clear that Alphabet is treating AI as an anchoring business strategy going forward.
While the public is familiar with the massive sums of money poured into AI, what these investments are focused on remains difficult for many to discern.
Google in particular has various enterprises focused on AI development, like Google DeepMind, which have created breakthroughs like Gemini and AlphaFold. Nevertheless, diverse teams conducting both theoretical and practical AI research have not gotten the recognition they may ultimately deserve.
Paradigms of Intelligence
A notable example of one of these research teams is Paradigms of Intelligence (Pi) at Google. As for what they do, their Github page states, “(Pi) brings together an interdisciplinary group of world-class researchers, engineers, and philosophers to explore the fundamental building blocks of intelligence and the conditions under which it can emerge.”
In exploring the nature of intelligence itself, their aim is to deepen our collective understanding of AI, which will allow them to develop AI that is more efficient and adaptable.
To learn more about the team and the profound nature of their research, I had the privilege of speaking with Mr. Yul Kwon of Google.
Kwon, who has worked for Google in a variety of different roles, is currently the VP/GM of Paradigms of Intelligence (Pi). He co-leads the team alongside Blaise Agüera y Arcas, CTO of Technology & Society at Google and the team’s founder. Kwon directly helps him bring together this versatile group of researchers.
Though he is not speaking on behalf of Google, throughout our conversation Kwon referred to AI’s rise as the most transformative technology that we have ever seen. He referenced how most major AI research labs are focused on building foundation models, training data, and leveraging immense computing power on a potential path towards an autonomous AI that could exceed human cognitive abilities (AGI). While this remains hypothetical, it does seem to be imminent.
Despite these rapid advancements, Kwon and his team are still inspired by the fact that the most intelligent artifact in the known universe is still the human brain. It is capable of astounding feats of intelligence that machines are only beginning to approach, and according to Yul, there are still fundamental insights about the nature and origins of intelligence that we can learn from and apply.
Perhaps most importantly, Google has provided the intellectual freedom to pursue these ideas and “explore big swings.”
Any Profound Insights?
Early on, Kwon described the Pi research team as “the most interesting group of people and work that you could potentially find on the planet.” Naturally, the scope of their research is astoundingly broad, which has led to research in a variety of fields.
Many examples were referenced, focusing notably on insights gathered from evolutionary biology. Kwon weaved through contrasting core Darwinian concepts like random mutations and natural selection with different evolutionary processes (including symbiogenesis, which emphasizes cooperation and interdependent merging rather than competition). It was clear to him that findings from evolutionary biology and other sciences inform the kinds of experiments humans do with computers.
Due to this, the team is actively working to try and come up with a more mathematical and theoretical framework, demonstrating how insights from diverse sciences might be useful in a computational context.
Kwon also explained how Blase Agüera y Arcas and the team have spearheaded the notion that computation is something far more fundamental than engineering, it is fundamental to both intelligence and life itself.
While one may not initially expect computer science to be intertwined with biology, understanding how intelligence works necessitates breaking down the walls of different disciplines. Kwon remarked that Agüera y Arcas was actually working on a mathematical framework for life and biology, a construct that could explain how evolution occurs.
It is these insights that may allow for a greater understanding of the process of self-replication in programs and how intelligence emerges from it. Despite this, Kwon maintains there is not one single science that can describe the vastness of the entire project.
Current AI Developments
Before Facebook, Google, and Silicon Valley, Yul Kwon worked in public service. With an impressive early resume that included McKinsey, a win on Survivor, and the FBI Academy, he joined the Obama-era FCC hoping to have a positive societal impact. However, Kwon quickly realized that technology, and by extension large tech companies like Google, are one of the most effective levers for driving positive societal impact at scale; as they are “building products that add value to people’s lives”.
His current work within the Pi research team has clearly deepened this belief, as Kwon argues that “we are on the cusp of something that could solve fundamental problems we have grappled with for our entire existence.”
Specifically, Kwon revealed how the Pi team is focused on distinct paradigm shifts enabled by AI development. For instance, with digital workflows now consisting of multi-agent societies, and different AI agents interacting with each other, the team is focused on whether this leads to a more efficient approach to collective intelligence.
Collective intelligence is intertwined with collaboration, arising from the fact that in nature and in human societies you have multiple agents and intelligences that work together (through division of labor and coordination). Collective intelligence far exceeds the individual intelligence of any one organism.
Kwon mentioned that there is a great deal of investigation on that front; in trying to understand the conditions under which AI agents might bend towards cooperation and pro-social behavior. On the other hand, if agents compete against each other there would be negative outcomes for the entire ecosystem. Ultimately, this is basic game theory.
Can we set the conditions for a hybrid society in which humans and AI agents work towards mutual flourishing as opposed to competition?
This is what most motivates Kwon, the conditions under which we develop artificial intelligence in a way that generates positive externalities, rather than one that brings about the oft-repeated AI doomsday scenarios.
Fears over AI
When I referenced some mainstream concerns over AI, Kwon noted that any claims of AI’s impact as being either categorically positive or negative aren’t likely to be helpful. “The reality is no one really knows”.
As with any new technology, Kwon understands that AI has the dual capacity to have both enormous benefits and detriments. Still, Kwon remains cautiously optimistic, with a healthy level of concern.
Even if the ultimate impact of AI is a net positive, there is a potential for significant disruptions along the way. In order to ensure the ideal outcome and maximize the benefits of AI, Kwon believes that multiple stakeholders in the public and private sectors would be required to work together in a collaborative fashion, to guide development and application of this technology.
While Kwon knows that competition not only exists between AI companies, but on a geopolitical scale, he deeply hopes that society will be able to identify potential risks and try to solve them constructively.
Ultimately, this is what motivates his interest in the Paradigms of Intelligence (Pi) team. Much of what Pi aims to do is understand the conditions under which intelligence arises. The better a society can understand these conditions, the better we can anticipate when tipping points occur.
Responsible Evolution
In mentioning regulation, Kwon noted that this is one of the different approaches and tools that a society can use to guide the responsible evolution of AI.
For its part, the Pi team is undertaking research that would promote cooperation and positive outcomes with and among AI agents, even in the absence of direct regulation. This is one area where they all hope to make a sizable contribution.
Can we develop tech so it bends towards positive, constructive pro-social applications?
In Kwon’s view: “it is unclear which direction things will go.. It depends on how we collectively work together to guide the development of this technology.”
Hopefully, alongside the Pi team, we can all strive to keep this massive ship sailing in the right direction.
Take-Home Points
- Major tech companies are investing heavily in AI, but the public may not fully understand what those investments are focused on.
- Google’s Pi team studies the foundations of intelligence to better understand how AI can become more efficient and adaptable.
- Insights from fields like evolutionary biology, computation, and mathematics may help explain how intelligence and self-replication emerge.
- AI agents could create stronger collective intelligence if they are designed to cooperate rather than compete.
- AI’s future impact is uncertain because it could bring both major benefits and significant disruptions.
- AI’s direction depends on how society collectively guides its development toward responsible, constructive, and pro-social outcomes.